Zhipu AI Eyes RMB 15B STAR Market Raise in China’s First Pure-Play LLM Listing Bid

Beijing-based Zhipu AI clears a critical regulatory hurdle for its Shanghai STAR Market listing, seeking RMB 15 billion (US$2.08 billion) to fund next-generation model development — even as annual net losses surpassed RMB 4.7 billion in 2025, exposing the razor-thin margin between hypergrowth and commercial viability.

The China Securities Regulatory Commission's Beijing bureau updated its guidance verification platform on June 17, 2026, confirming that Beijing Zhipu Huazhang Technology, commonly known as Zhipu AI, has formally completed its IPO counseling inspection — the final administrative checkpoint before submitting a prospectus to the Shanghai Stock Exchange. Guotai Haitong Securities serves as the sole counseling institution for this stage.

The milestone arrives less than six months after Zhipu AI listed on the Hong Kong Stock Exchange on January 8, 2026 (02513.HK) at an initial market capitalization exceeding HK$51 billion, making it the world's first independent general-purpose large language model (LLM) company to achieve public listing. The accelerated pivot back to the A-share market — Zhipu withdrew its original counseling filing in February 2026 and restructured the arrangement with Guotai Haitong and China International Capital Corporation — signals that management views domestic capital as structurally indispensable, not merely supplementary.

Fundraising Structure Reveals Where the Burn Rate Is Going

On June 1, 2026, Zhipu's board approved a STAR Market issuance plan targeting a 2%–8% free float, with a projected gross raise of approximately RMB 15 billion (US$2.08 billion). The capital allocation is surgically focused:

• RMB 12 billion (US$1.67 billion) — next-generation GLM-series foundational model R&D, covering pretraining architecture upgrades and compute cluster expansion
• RMB 2 billion (US$278 million) — MaaS (Model-as-a-Service) platform upgrades, developer ecosystem buildout and commercialization infrastructure
• RMB 1 billion (US$139 million) — working capital to sustain R&D operations and cash flow

The 80% allocation to raw compute and model iteration is not incidental. In 2025, Zhipu's R&D expenditure reached RMB 3.18 billion (US$442 million), up from RMB 2.195 billion (US$305 million) in 2024, with GPU procurement constituting the dominant cost line. The STAR Market raise is, in effect, a structured bridge to the next model generation before cloud-side unit economics improve sufficiently to self-fund at scale.

Revenue Compounding at 130%-Plus, But Losses Accelerate Faster

Zhipu's top-line trajectory is among the most aggressive in China's enterprise AI sector. Revenue grew from RMB 57.41 million in 2022 to RMB 125 million in 2023, RMB 312 million in 2024, and RMB 724 million (US$100.6 million) in 2025 — a three-year compound annual growth rate of approximately 130%, with the 2025 year-on-year growth rate holding at 131.9%.

Yet the loss curve has steepened more sharply. Net losses widened from RMB 144 million in 2022 to RMB 788 million in 2023, RMB 2.958 billion in 2024, and RMB 4.718 billion (US$655 million) in 2025. The divergence between revenue and loss trajectories reflects a deliberate preemptive investment posture — but it also means that at current burn rates, Zhipu's existing liquidity runway is finite without external capital infusions.

The 2025 gross margin of 41.0% sits within a reasonable range for enterprise software-adjacent businesses, but masks a structural asymmetry: the private deployment segment, which contributed RMB 534 million (73.7% of total revenue) in 2025, carries materially higher margins than the cloud-based MaaS segment. The MaaS business, while growing at 292.6% year-on-year to RMB 190 million in 2025, remains a margin drag at current scale.

Private Deployment Anchors Revenue While Cloud Business Chases Scale

Zhipu's business model bifurcates into two strategically distinct segments. The private deployment arm serves central state-owned enterprises, financial institutions, and energy groups — clients with stringent data sovereignty requirements — achieving a contract renewal rate of 95%. This segment functions as a high-visibility, high-retention revenue base.

The cloud MaaS platform, by contrast, is the growth optionality bet. As of Q1 2026, annualized recurring revenue (ARR) from the MaaS platform reached RMB 1.7 billion (US$236 million), reflecting the compounding effect of Zhipu's developer ecosystem, which had surpassed 45 million registered developers and served over 12,000 enterprise clients globally as of September 2025.

The GLM-series models — covering natural language, code generation, multimodal outputs, and agentic applications — are fully proprietary, including pretraining frameworks and core algorithms. This full-stack independence enables compatibility with domestic compute infrastructure and the broader "Xinchuang" technology substitution ecosystem, a critical differentiator when competing for government and SOE contracts against cloud hyperscalers such as Baidu, Alibaba, and ByteDance.

Market Structure Shifts From Model Racing to Monetization Discipline

According to Frost & Sullivan data cited in Zhipu's prospectus, China's large language model market reached RMB 5.3 billion in 2024 and is projected to expand to RMB 101.1 billion (US$14.04 billion) by 2030, implying a CAGR of 63.7%. Enterprise demand is expected to account for approximately 90% of that market, driven by government digitalization mandates, vertical-industry model customization, and AI agent deployment.

The competitive topology has evolved from the "hundred-model wars" of 2023–2024 into a bifurcated oligopoly: internet conglomerates dominating general-purpose cloud inference on the back of infrastructure scale and consumer traffic, while independent vendors like Zhipu carve defensible positions in private deployment through technical neutrality and regulatory alignment.

However, the risk matrix is non-trivial. Internet platforms are aggressively cutting API pricing, compressing cloud-side margins across the board. Established vertical technology vendors are intensifying competition for government and SOE contracts. Geopolitical constraints on high-end GPU supply chains introduce procurement uncertainty. And Zhipu's top-client revenue concentration remains a disclosure risk for prospective A-share investors.

STAR Market Listing Unlocks Domestic Capital, But Path to Profitability Remains the Central Question

The strategic logic of a dual Hong Kong–STAR Market structure is transparent: the STAR Market provides access to domestic institutional and retail capital that cannot participate in Hong Kong-listed securities, broadens Zhipu's brand visibility among the government and SOE client base it depends on, and allows the company to capture the "hard tech" valuation premium embedded in STAR Market multiples.

Zhipu was founded in 2019 as a commercialization spinout from Tsinghua University's Knowledge Engineering Group (KEG), led by Professor Tang Jie. It holds a 6.6% share of the domestic independent general-purpose LLM vendor market by 2024 revenue — the highest among pure-play peers.

The completion of the STAR Market counseling inspection positions Zhipu to submit its formal prospectus to the Shanghai Stock Exchange imminently. If approved, it would become the first pure-play general-purpose LLM company listed on the A-share market, filling what analysts have described as a structural gap in China's domestic AI equity universe.

The fundamental question facing STAR Market investors, however, is the same one that has shadowed every LLM platform globally: whether the path from triple-digit revenue growth to sustainable unit economics can be navigated before the capital markets window narrows. For Zhipu, the answer will hinge on accelerating the mix shift toward standardized cloud products, compressing per-unit compute costs, and converting its developer ecosystem into recurring enterprise revenue at a pace that outstrips the loss curve.

Related Coverage:

Zhipu AI Shares Surge 33% After U.S. Export Controls Ground Anthropic's Latest Models